Neural embeddings of transaction data

Abstract

Systems, methods, and computer program products to provide neural embeddings of transaction data. A network graph of transaction data based on a plurality of transactions may be received. The network graph of transaction data may define relationships between the transactions, each transaction associated with at least a merchant and an account. A neural network may be trained based on training data comprising a plurality of positive entity pairs and a plurality of negative entity pairs. An embedding function may then encode transaction data for a first new transaction. An embeddings layer of the neural network may determine a vector for the first new transaction based on the encoded transaction data for the first new transaction. A similarity between the vectors for the transactions may be determined. The first new transaction may be determined to be related to the second transaction based on the similarity.

What is claimed is: 
     
       1. A system, comprising:
 a processor circuit; and 
 a memory storing instructions which when executed by the processor circuit cause the processor circuit to:
 receive a network graph of transaction data based on a plurality of transactions, the network graph defining relationships between the plurality of transactions, each transaction associated with at least one account of a plurality of accounts; 
 select training data comprising a plurality of positive entity pairs from the network graph of transaction data and a plurality of negative entity pairs not present in the network graph of transaction data, the negative entity pairs comprising artificially generated relationships between each entity in the negative entity pair, wherein the training data is selected based on a respective time between a respective timestamp of the transactions of each positive entity pair being less than a time threshold; and 
 train a neural network based on the training data, the neural network comprising an embeddings layer. 
 
 
     
     
       2. The system of  claim 1 , the memory storing instructions which when executed by the processor circuit cause the processor circuit to:
 train a model based on the embeddings layer of the neural network, the model associated with each of the plurality of accounts. 
 
     
     
       3. The system of  claim 1 , the memory storing instructions which when executed by the processor circuit cause the processor circuit to:
 receive a transaction log; and 
 generate the network graph of transaction data based on one or more extract transform load (ETL) operations applied to the transaction log. 
 
     
     
       4. The system of  claim 3 , the one or more ETL operations comprising: (i) standardizing the transaction log according to one or more formats, and (ii) assigning a unique identifier to each unique customer account in the transaction log. 
     
     
       5. The system of  claim 4 , the memory storing instructions which when executed by the processor circuit cause the processor circuit to:
 select a predefined number of the positive entity pairs from the network graph of transaction data; 
 generate, for each positive entity pair, a predefined number of negative entity pairs; and 
 generate the embeddings layer comprising a plurality of embedding values based on the training of the neural network using the selected positive entity pairs and the generated negative entity pairs, wherein the embeddings layer associates each embedding value with one of the unique identifiers. 
 
     
     
       6. The system of  claim 1 , wherein a first negative entity pair of the plurality of negative entity pairs comprises a first transaction with a first merchant and a second transaction with a second merchant, the first negative entity pair selected based on the time between the timestamps of the first and second transactions being greater than the time threshold. 
     
     
       7. The system of  claim 1 , wherein the training of the neural network refines a plurality of embedding values of the embeddings layer such that a respective distance between each positive entity pair is minimized relative to initial embedding values for the embeddings layer and a respective distance between each negative entity pair is minimized relative to the initial embedding values for the embeddings layer. 
     
     
       8. A non-transitory computer-readable storage medium storing instructions that when executed by a processor of a computing device, cause the processor to:
 receive a network graph of transaction data based on a plurality of transactions, the network graph defining relationships between the plurality of transactions, each transaction associated with at least one account of a plurality of accounts; 
 select training data comprising a plurality of positive entity pairs from the network graph of transaction data and a plurality of negative entity pairs not present in the network graph of transaction data, the negative entity pairs comprising artificially generated relationships between each entity in the negative entity pair, wherein the training data is selected based on a respective time between a respective timestamp of the transactions of each positive entity pair being less than a time threshold; and 
 train a neural network based on the training data, the neural network comprising an embeddings layer. 
 
     
     
       9. The non-transitory computer-readable storage medium of  claim 8 , further storing instructions that when executed by the processor cause the processor to:
 train a model based on the embeddings layer of the neural network, the model associated with each of the plurality of accounts. 
 
     
     
       10. The non-transitory computer-readable storage medium of  claim 8 , further storing instructions that when executed by the processor cause the processor to:
 receive a transaction log; and 
 generate the network graph of transaction data based on one or more extract transform load (ETL) operations applied to the transaction log. 
 
     
     
       11. The non-transitory computer-readable storage medium of  claim 10 , the one or more ETL operations comprising: (i) standardizing the transaction log according to one or more formats, and (ii) assigning a unique identifier to each unique customer account in the transaction log. 
     
     
       12. The non-transitory computer-readable storage medium of  claim 11 , further storing instructions that when executed by the processor cause the processor to:
 select a predefined number of the positive entity pairs from the network graph of transaction data; 
 generate, for each positive entity pair, a predefined number of negative entity pairs; and 
 generate the embeddings layer comprising a plurality of embedding values based on the training of the neural network using the selected positive entity pairs and the generated negative entity pairs, wherein the embeddings layer associates each embedding value with one of the unique identifiers. 
 
     
     
       13. The non-transitory computer-readable storage medium of  claim 8 , wherein a first negative entity pair of the plurality of negative entity pairs comprises a first transaction with a first merchant and a second transaction with a second merchant, the first negative entity pair selected based on the time between the timestamps of the first and second transactions being greater than the time threshold. 
     
     
       14. The non-transitory computer-readable storage medium of  claim 8 , wherein the training of the neural network refines a plurality of embedding values of the embeddings layer such that a respective distance between each positive entity pair is minimized relative to initial embedding values for the embeddings layer and a respective distance between each negative entity pair is minimized relative to the initial embedding values for the embeddings layer. 
     
     
       15. A method, comprising:
 receiving, by a computer processor, a network graph of transaction data based on a plurality of transactions, the network graph defining relationships between the plurality of transactions, each transaction associated with at least one account of a plurality of accounts; 
 selecting, by the processor, training data comprising a plurality of positive entity pairs from the network graph of transaction data and a plurality of negative entity pairs not present in the network graph of transaction data, the negative entity pairs comprising artificially generated relationships between each entity in the negative entity pair, wherein the training data is selected based on a respective time between a respective timestamp of the transactions of each positive entity pair being less than a time threshold; and 
 training, by the processor, a neural network based on the training data, the neural network comprising an embeddings layer. 
 
     
     
       16. The method of  claim 15 , further comprising:
 receiving, by the processor, a transaction log; and 
 generating, by the processor, the network graph of transaction data based on one or more extract transform load (ETL) operations applied to the transaction log. 
 
     
     
       17. The method of  claim 16 , the one or more ETL operations comprising: (i) standardizing the transaction log according to one or more formats, and (ii) assigning a unique identifier to each unique customer account in the transaction log. 
     
     
       18. The method of  claim 17 , further comprising:
 selecting, by the processor, a predefined number of the positive entity pairs from the network graph of transaction data; 
 generating, by the processor for each positive entity pair, a predefined number of negative entity pairs; and 
 generating, by the processor, the embeddings layer comprising a plurality of embedding values based on the training of the neural network using the selected positive entity pairs and the generated negative entity pairs, wherein the embeddings layer associates each embedding value with one of the unique identifiers. 
 
     
     
       19. The method of  claim 15 , wherein a first negative entity pair of the plurality of negative entity pairs comprises a first transaction with a first merchant and a second transaction with a second merchant, the first negative entity pair selected based on the time between the timestamps of the first and second transactions being greater than the time threshold. 
     
     
       20. The method of  claim 15 , wherein the training of the neural network refines a plurality of embedding values of the embeddings layer such that a respective distance between each positive entity pair is minimized relative to initial embedding values for the embeddings layer and a respective distance between each negative entity pair is minimized relative to the initial embedding values for the embeddings layer.